goldberg phd 1st half

8/2/2019 Goldberg PHD 1st Half

1/133

Verb-Stranding VP Ellipsis: A Cross-Linguistic Study

by

Lotus Madelyn Goldberg

June 2005

Bachelor of Arts, Linguistics, with Honors in the Major, University of California at Santa Cruz (1993)Master of Arts, Linguistics, University of California at Santa Cruz (1998)

A thesis submitted toMcGill Universityin partial fulfilment of the requirements

of the degree ofDoctor of Philosophy

Department of LinguisticsMcGill UniversityMontral, Qubec, Canada Lotus Goldberg 2005


2/133

ii

Abstract

This thesis presents a study of a construction which I refer to as Verb-Stranding VP Ellipsis.

The construction is studied here, specifically, in two distinct senses.

First, in chapter two, diagnostics are proposed by which the VP Ellipsis (VPE) construction

can be identifiedirrespective of whether the main verb involved is null or overt. It is proposed that

these diagnostics can be used to rule out the possibility that the data at issue are cases of other types

of null anaphora, such as null arguments, Stripping, Gapping, and Null Complement Anaphora. It

emerges from this section of the thesis that Modern Hebrew, Modern Irish, and Swahili have V-

Stranding VPE data which form a natural class with English's Aux-Stranding VPE, while Japanese,

Korean, Italian, and Spanish do not.

The second focus is the question of how V-Stranding VPE should be generated. Chapters 3

and 4 argue in favor of an analysis involving PF Deletion of a VP out of which the main verb has

raised, and against an LF Copying treatment. These arguments, in part, involve the Verbal Identity

Requirementon VP Ellipsis, a novel generalization involving strict identity in root and derivational

morphology between the antecedent- and target-clause main Vs of the construction. Within the

previously known requirement that elided phrases express semantically Given information, I argue

that this generalization results from the fact that the head of an elided phrase must itself express Given

informationwhether or not the head surfaces as phonologically null.


3/133

iii

Rsum

Dans cette tude, on considre en dtail une construction que j'appelle L'lision d'une

expression verbale sans l'lision du verbe principal (anglais V-Stranding VP Ellipsis ). Cette

construction est tudie ici, spcifiquement, dans deux sens distincts.

Dans le chaptre 2, on propose des diagnostics grce auxquels on peut identifier la

construction lision d'une expression verbale ( EEV , anglais VP Ellipsis ), que le verbe

principal dans l'expression verbale soit manifeste ou lid. On soutient que ces diagnostics peuvent

tre utiliss pour liminer la possibilit que les donnes pertinentes soient des exemples d'autres types

d'anaphore nulle, tels que argument du verbe nul, le Stripping , le Gapping , et le Null

Complement Anaphora . Ainsi, on propose dans cette section que l'EEV sans llision du verbe

dans les grammaires de l'hbreu, de l'irlaindais et du swahili forme une classe naturelle avec lEEV

avec llision du verbe en anglais. On soutient aussi que cette construction n'xistent pas en japonais,

en coren, en espagnol, ou en italienne.

Ensuite, on considre la question de comment gnrerles exemples d'EEV sans llision du

verbe. Dans les chaptres 3 et 4, on propose une analyse qui utilise la suppression d'une expression

verbale au niveau de la Forme Phonologique ( la suppression FP , anglais PF Deletion ) aprs

le dplacement du verbe principal une position en dehors de l'expression verbale, et on prsente une

explication de la raison pour laquelle une analyse qui utilise des copies de la Forme Logique ( copie

FL , anglais LF Copying ) n'est pas viable. Ceci implique, en partie, la Condition d'Identit

Verbale, une gnralisation propose ici pour la premire fois, impliquant une identit stricte de la

racine et dans la morphologie drivationnelle entre les verbes principaux des propositions


4/133

iv

antcdentes et des propositions cibles. Dans le cadre de la condition connue selon laquelle les

syntagmes lids expriment une information smantique donne (anglais Given ), je soutiens que

la condition d'identit verbale rsulte du fait que la tte d'un syntagme lid doit elle-mme exprimer

l'information donne smantiquementque la tte soit phonologiquement manifeste ou nulle.


5/133

v

Acknowledgements

The number of people who have helped me get to this point, and helped me bring this thesisinto existence, is remarkable.

My deepest thanks go to my supervisor, Lisa deMena Travis. She has been a source ofsupport and encouragement from the start of my time at McGill, and has been an anchor for me asI struggled to work on and complete this piece of work. Her insights into the workings of headmovement, and her intuitions into the underlying workings of the languages and data involved, havebeen invaluable during this entire process. It is an understatement to say that I would not have madeit to this point without her.

I owe much as well to Jason Merchant, who generously agreed to serve on my thesiscommittee, despite the fact that he is at the University of Chicago. During both my time in Montraland the year I spent in Chicago, he has been an invaluable source of knowledge and information. Hisencouragement and supportand his belief that this project was worthwhile and importanthavebeen invaluable as well.

My thanks also to Nigel Duffield, the third member of my committee, to Brendan Gillon, PeterHallman and Jon Nissenbaum, who served on my defense committee, and to Norvin Richards and JimMcGilvray, who served as external examiners. All have given me good points to think about for thetime to come.

This dissertation's existence and approach owe a fundamental debt to the questions posed tome by Jim McCloskey and Jorge Hankamer at the oral defense of my UC Santa Cruz MA thesis. Andmy own existence and approach as a linguist and academic owe just as much to education and supportthat they, Sandy Chung, and Judith Aissen have given me. Their work with me during my time at

UCSC (along with that of the other faculty and community of the department) went far beyondanything that a BA and then MA student could have imagined. Their presence and guidance after Ileft has been an extraordinary gift.

The tack taken by this thesis and in my earlier, related research on Hebrew Null Objects owesa great debt to the initial study of these phenomena carried out by Edit Doron (1990, 1999). It isnoteworthy that the profile of both Hebrew VP Ellipsis and Hebrew Null Objects which emerges hereis entirely consistent with the majority of claims originally made in Doron (1990).

Portions of the material in this thesis have also benefitted from comments and discussion withlinguists outside my committee. I am particularly grateful to Kyle Johnson, Chris Kennedy, Jeff Lidz,Line Mikkelsen, Gregory Ward, and, most especially, to Joey Sabbagh and to The Task Master. My

deepest thanks as well to Beverly Bouwsma of Berkeley, California, for the wonderful gift of a quietand sane place to work during this projects last and most difficult stages.


6/133

Goldberg

vi

I very gratefully acknowledge the time, patience, and insights contributed by the native Hebrewspeakers with whom I have worked during my time at McGill. All new Hebrew data included herewere elicited from them, and neither this thesis nor the other work on Hebrew which I completed at

McGill would exist without them. They include Yifat Mamut, Itsik Romano, Tal Shor, Shiri Ofir,Carmela Aigen, Chava Dienar, Roy Chorev, Hidai Friedman, Cigal Gabbay, Rony Greenberg, andZohar Mamut.

This thesis owes an enormous debt to the staff of McGill's McLennan-Redpath InterlibraryLoan office, who handled my endless requests with remarkable speed and good humor. Without theirefforts, the scope for which I have aimed would have remained far beyond my reach. Finally, Igratefully acknowledge the support of portions of this thesis and of some of my earlier McGill workfrom SSHRCC research grants 410-93-0897 to Lisa deMena Travis and 410-99-0902 to JonathanDavid Bobaljik.

Turning to my student life outside the dissertation itself, I am grateful to the larger communityof McGill faculty (past and present) for their courses, discussions, and support of all kinds: JulieAuger, Mark Baker, Jonathan Bobaljik, Charles Boberg, Veena Dwivedi, Brendan Gillon, HeatherGoad, Yosef Grodzinsky, Eva Kehayia, Michel Paradis, Glyne Piggott, Lydia White, and SusiWurmbrand. Et, Michel, merci mille fois pour lassistance avec le rsum!

Although Mark Baker left McGill shortly after I arrived here, he has remained a source ofinsight and encouragementparticularly during a meeting with him at a crucial moment for me duringthe summer of 2001. I thank Jonathan Bobaljik for the time and effort which he invested into myearlier work on Hebrew Null Objects. Some of the content from that work is included in Chapter 2in more developed form, and much of the thinking which I did during that time has influenced myapproach here in obvious ways. My thanks also to Brendan Gillon for his insights into logic and

semantics, and for being consistently willing to discuss various theoretical viewpoints andmethodological issues with me and others in the department. And lots of thanks to Glyne Piggott andLydia White for their assistance with and support of my departmental and teaching progress andduring my time here.

My thanks to Lise Vinette, Andria De Luca, and Linda Suen for assistance with all aspects ofmy departmental life.

Outside McGill, I am grateful to various other Montral linguists for their support andkindness during my time here, including Claire Lefebvre and Ayumi Matsuo. I owe a special debt toDana Isaac, Mark Hale, and, especially, Charles Reiss, for making the linguistics world at ConcordiaUniversity such a stimulating and welcoming second place for me and other McGill students topresent work and hear about the work of others. Evenings at Charles and Dana's house were alsoa highlight for me during my last year in Montral.


7/133

Goldberg

vii

The community of linguistics graduate students at McGill has been a constant andextraordinary source of support, motivation, information, discussion, and friendshipas well asfellow consumers of an awful lot of dessert, coffee, and beer. I dont have words to express the

depth of my feelings for those to whom I have been especially close, so I will just list them here, thankthem for everything, and send out to them all the warmth, good wishes, and hugs that they can bear:Hiro Hosoi, Mika Kizu, Ingrid Leung, Mikinari Matsuoka, Corrine McCarthy, Heather Newell,Naoko Tomioka, and Mikael Vinka. Ditto, outside McGill, and for more than I can describe, to JoeySabbagh, Alan Yu, Patrick Davidson, and Line Mikkelsen. And ditto again, outside linguistics, tomy friends from California and their families: Theresa Vernetti, Laurie Wilson, Bonnie BrownGregory, Rebecca Holder MacDonald, Erin Koch, and Leslie Teruya.

I also acknowledge and thank the other graduate students who have come through ourdepartment during my time here; there isnt one whom I have not benefitted from knowing. Lots ofgood wishes and thanks to especially Ken Ariji, Alan Bale, Hela Ben Ayed, Walcir Cardoso, Simone

Conradie, Mario Fadda, Theres Grter, Aye Grel, Hyun-Sook Kang, Yuhko Kayama, KaterinaKlepousniotou, Erika Lawrence, John Lewis, Mizuki Mazzotta, Evan Mellander, Jennifer Mortimer,Junko Murai, Tomoyo Oda, Ileana Paul, Asya Pereltsvaig, the late and very dearly missed LaraRiente, Yvan Rose, Pablo Ruiz, Jeff Steele, Tomokazu Takehisa, Hidekazu Tanaka, Tohru Uchiumi,Monica Ungureanu, Elena Valenzuela, Teresa Ching-Huei Wu, and Myunghyun Yoo.

My deepest thanks, for more than I can name, to my parents, Russell and Joan Goldberg, andmy brother, Henry Goldberg. I hope they know how much I love them. Within my larger extendedfamily, deepest thanks and love too to especially my aunt and uncle Lois and Arthur Goodman.

I also want to acknowledge the love, support, and time given to me by my grandparents,Milton and Frances Brown, and Mack and Lenore Goldberg, which has contributed so strongly to

the person I have become. I know that I would never have had the opportunity to choose this courseof life had it not been for the sacrifices which they and their parents made, and for this too I willalways be deeply grateful. Although I regret that they will not be able to read this, I hope that itwould have made them proud.

Warmest thanks and affection, finally, to the members of the Sciuridfamily living on andaround Mount Royal, who kept me such good company during my time in Montral. They, ofcourse, could not care less about any of thiswhich must be at least part of the reason that they areso charming.


8/133

viii

For Jerry,

with my love and profound gratitude


9/133

ix

Table of Contents

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii

Rsum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii

Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v

Chapter 1: Introduction

1. Introduction: Core Data Considered in this Thesis . . . . . . . . . . . . . . . . 1

2. Key Aims of the Thesis, and a Chapter-by-Chapter Summary . . . . . . . . 5

2.1. Overview of the key aims and tasks undertaken . . . . . . . . . . . . . . 5

2.2. Task #1: Establishing a V-Stranding VP Ellipsis diagnosis . . . . 6

2.3. Task #2: Bringing V-Stranding VPE data to bear on the theory

of VP Ellipsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103. On the Motivation for Integrating V-Stranding VP Ellipsis Data into

(VP) Ellipsis Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

Chapter 2: Establishing a Diagnosis of V-Stranding VP Ellipsis

0. Introduction and Chapter Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

1. A First Consideration of VPE Diagnostics . . . . . . . . . . . . . . . . . . . . . . . 27

1.1. Licensing by an inflectional head . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.2. Additional traits characteristic of English VPE . . . . . . . . . . . . . . 31

2. The Null Argument Versus VP Ellipsis Ambiguity Problem, and InitialIssues in Controlling for Null Argument Structures . . . . . . . . . . . . . . . . 35

2.1. The problem of ambiguity between a VPE and null argumentanalysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

2.2. Initial control techniques for null argument structures inV-Stranding VPE examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3. Teasing Null Objects Apart from VP Ellipsis in Hebrew . . . . . . . . . . . . 42

4. Ruling Out a Null Object Analysis for Swahili and Ndendeule VPEData . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5. Ruling Out a Null Argument Analysis for Irish VPE Data . . . . . . . . . . . 58

5.1. The motivation for V-to-Infl raising in Irish . . . . . . . . . . . . . . . . . 59

5.2. The distribution of VP-internal null arguments in Irish . . . . . . . . 64

5.3. The existence of putative Irish VPE examples which cannot be


10/133

Goldberg

x

instances of independently null VP-internal arguments . . . . . . . . 72

6. The Case Against V-Stranding VP Ellipsis in Japanese and Korean . . . 73

6.1. Overview of the motivation for V-Stranding VPE in Japaneseand Korean . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

6.2. Japanese and Korean VP-internal elements can elide when the restof the VP is overt . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

6.3. The Otani and Whitman (1991) proposal . . . . . . . . . . . . . . . . . . . 82

6.4. Argument #1 against VPE: Japanese and Korean VP-adverbialscan elide neither independently nor in null VPs . . . . . . . . . . . . . . 89

6.5. Argument #2 against VPE: Japanese and Korean sloppy identityreading availability correlates neither with English VPE facts, nor

with a VPE structure being possible . . . . . . . . . . . . . . . . . . . . . . . 93

6.5.1. Case (Ai): Japanese lacks sloppy identity where EnglishVPE allows it . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

6.5.2. Case (Aii): Japanese allows sloppy identity where EnglishVPE does not . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

6.5.3. Case (Bi): Korean lacks sloppy identity where a VPEstructure is possible . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

6.5.4. Case (Bii): Japanese and Korean allow sloppy identitywhere a VPE structure is impossible . . . . . . . . . . . . . . . . 107

6.6. Argument #3 against the VPE Proposal: Sloppy identity readingscan be derived for null DPs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

6.7. Summing up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

7. Chapter Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120


11/133

Goldberg

xi

Chapter 3: Deriving the Verb Stranding Effect


1. The PF Deletion Analysis of Ellipsis . . . . . . . . . . . . . . . . . . . . . . . . . . . 1282. The LF Copying Analysis of Ellipsis . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

2.1. Wrapping up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

3. Deriving V-Stranding VPE via PF Deletion: Part one . . . . . . . . . . . . . . 141

4. The Argument Against LF Copying from V-Stranding VPE: Part one . 146


Chapter 4: Capturing the Isomorphism Requirements of V-Stranding VPE


1. The Verbal Identity Requirement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

1.1. The Verbal Identity Requirement in Hebrew . . . . . . . . . . . . . . . . 157

1.2. On the cross-linguistic generality of the Verbal IdentityRequirement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

1.3. The novelty of the Verbal Identity Requirement within the largerdomain of ellipsis isomorphism . . . . . . . . . . . . . . . . . . . . . . . . . . 171

2. The Proposed Account: Deriving the Verbal Identity Requirementwith PF Deletion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175

3. The Argument Against LF Copying: Part Two . . . . . . . . . . . . . . . . . . . 184


Chapter 5: Conclusion


1. Summary and Implications of the Arguments Made in this Thesis . . . . 200

2. Issues for Further Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204

2.1. Diagnosing additional cases of VP Ellipsis . . . . . . . . . . . . . . . . . 205

2.2. What is VP Ellipsis?: Taking stock of where we now stand . . . . 207

2.3. Analytic questions and predictions . . . . . . . . . . . . . . . . . . . . . . . . 2103. Final Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215


12/133

1This work is carried out within the generative, synchronic linguistic tradition: thus 'Hebrew' here refers just toModern Hebrew, 'Irish' refers just to Modern Irish, and so on.

2Conventions used in examples throughout this thesis are as follows. First, Hebrew examples are presented withNorth American transcription conventions; additionally, [c] represents the affricate [ts].

The bracketing of the internal arguments in English translations of V-Stranding VPE examples is used just as anotational device, to group together the elements which lie internal to the targeted VP at the representational levelrelevant for VP Ellipsis.

(continued...)

1

Chapter 1

Introduction

1. Introduction: Core Data Considered in this Thesis

This thesis concerns itself with a type of Verb Phrase Ellipsis (VP Ellipsis or VPE) in

languages outside English, and in which the main verb (V) of the clause in which the elision occurs

(referred to here as the target clause) remains overt obligatorily, rather than eliding. A typical

example of the well-studied English VP Ellipsis construction appears in (1). Here, the main V of the

target clause obligatorily elides, along with the rest of the VP, leaving only one or more auxiliary V

overt:

English VPE with Main V ObligatorilyNull

(1) Arthur [VP brought a present to Hall],and Julia did [bring a present to Hall] too* and Julia brought too; *and Julia will bring too.

In contrast to such English examples, I will be concerned almost solely here with non-English data

such as (2) from Hebrew, (3) from Irish, and (4) from Swahili.1 These examples differ superficially

from their English counterparts in displaying an overttarget-clause main V which bears all verbal

inflection, thus lacking auxiliary Vs altogether. For this reason, I use the term V(erb)-Stranding VP

Ellipsis to describe this type of VPE:2


13/133

Goldberg, Chapter 1 Introduction 2

2(...continued)In Hebrew morpheme glosses, the traditionally-termed Benoni conjugation used in Hebrew present tense and

participial verb forms is glossed as [Bni]. Thus, yoda'at, the present tense feminine 'knows' as in 'she knows', isglossed as know[BniFsg]. Morpheme glosses abstract away from certain aspects of morphological detail when theyare not relevant to issues at hand; thus, Hebrew li, for instance, is generally glossed as 'to.me' rather than as the morestrictly correct 'DAT[1sg]'.

As standard in the literature, Irish data are presented here using the language's orthography; see e.g. Siadhail(1989) for pronunciation guides. Irish VPE examples are sometimes given with more literal, pseudo-Englishtranslations which preserve Irish's VSO finite clausal basic order. This is done to make more transparent the fact thatthe post-verbal subject is among the VPE-internal elements which are elided. Chapter 2 presents explicit discussionof this fact, as well as of the fact that such elided subjects are notinstances of Irish's subjectpro when shown as struck-

through, since they coincide in such cases with analytic (as opposed to synthetic) V forms devoid of subject agreementmorphology. To make the latter fact more transparent, Irish analytic V forms are glossed using the abbreviation Anl;thus, cuireann s, 'put[PresAnl] he', i.e. 'he puts'. In contrast, synthetic V forms can be identified by the presence ofsubject agreement in their glosses, e.g. cuirim, 'put[Pres1sg]', i.e. '(I) put'.

'FV' in Bantu morpheme glosses represents a 'Final Vowel'. Note that I will remain neutral about whether theobject clitic marked on verbs is ultimately best analyzed an object-V agreement morpheme, as argued by e.g. Bergvall(1986), Kinyalolo (1991), and Ngonyani (1998), or, rather, as an incorporated object pronoun, as in Bresnan andMchombo (1987) and Bresnan (1993). The choice between the two analyses will not crucially affect the tenability ofthe VPE analysis argued for here for Swahili and Ndendeule examples, so far as I can see. However, for the sake ofconcreteness, I will refer to this element as an object agreement morpheme in the present text.

Hebrew VPE with Main V Obligatorily Overt (Doron 1999:ex.13)

(2) Q: alaxt etmol et ha-yeladim le-beit-ha-sefer?send[Past2Fsg] yesterday ACC the-children to-house-the-book

'(Did you) send [yesterday the children to school] ?'

A: alaxti.send[Past1sg]

'(I) sent [yesterday the children to school] .'

(cf. English '...I did [send the children to school yesterday].')

Irish VPE with Main V Obligatorily Overt (McCloskey 1991:ex.27a)

(3) Dirt m go gceanninn agus cheannaigh.said I COMP buy[Condit1sg] it and buy[PastAnl]

lit. 'Said I that would buy [(I) it] and bought [I it]. '(cf. English '...and I did [buy it].')


14/133


3'IP' and 'VP' are used throughout this thesis, in trees as well as text, as abbreviations for the entire inflectionaland verb-phrasal domains, respectively, wherever the exact identity of the subparts of each is not at issue. (Thus, undercurrent standard conceptions of phrase structure, 'VP' will often indicate e.g. vP or voiceP, and so on.) Note too thatI have chosen to use X-bar theory in trees here, but, as far as I have been able to tell, nothing crucial hinges on thismove.

Swahili VPE with Main V Obligatorily Overt (Ngonyani 1996a:ex.20a)

(4) Mama a-li-tak-a ku-m-nunul-i-a m-toto vi-atumother 1Su-Past-WANT-FV Infin-1Obj-BUY-Applic-FV 1-child 8-shoe

na baba a-li-tak-a pia.and father 1Su-Past-WANT-FV also

'Mother wanted [to buy the child shoes] and father wanted [to buy the child shoes]too.'

(cf. English '...and father did [want to buy the child shoes] too.')

For these languages, such data were first argued to involve a variety of VP Ellipsis by Doron

(1990, 1999), Ussushkin/Sherman (1997), Goldberg (1998, 2001) for Hebrew; McCloskey (1991,

1995) for Irish; and Ngonyani (1996a,b, 1998) for Swahili and Ndendeule. For each language, these

researchers argued the effect of main V stranding to arise because there is independent motivation

that the main V lies in a high position, within the inflectional domain, in overt syntax.

Within the Chomskyan approach in which this thesis is written, this high verbal position is

traditionally viewed as the result of obligatory main V raising to (a subpart of) Inflthe inflectional

layer of the clause which lies immediately above the verb phrase. V-Stranding VPE examples like

those in (2-4) are thus argued, on this view, to involve ellipsis of a VP whose main V actually lies in

a position which c-commands the null constituent. The material which lies internal to the VP at the

level of representation relevant for VP Ellipsis will therefore exclude the main V, with the effect that

the main V is left overt. This is shown schematically as follows:3


15/133


VP Ellipsis Schematic for V Raising Languages: Main V Lies outside the VP(5) IP

3

I' 3

[Imain-V] VP6

....tVerb....

V-Stranding VP Ellipsis of this sort has now been referred to as existing in sixteen different

languages within the extant literature, all such claims having emerged relatively recently, since roughly

the early 1990s. The extent to which the different languages involved have been accepted by the field

as a whole as having this construction has varied a great deal. Much of this literature, furthermore,

has so far escaped the attention of many linguists, as discussed further in Section 3 below.

Nonetheless, such V-stranding examples for Hebrew, Irish, and Swahili are motivated

extensively in Chapter 2 to instantiate the same basic VP Ellipsis phenomenon as is familiar from

English. Most significantly, on this point, the discussion there argues that the target clauses

instantiating VP Ellipsis in these languages appear in the same broad range of sentence types as does

English VP Ellipsis. Such sentence types include coordinated CPs, adjacent CPs uttered by the same

speaker, and question-answer pairs that cross a speaker boundary; targeted null VPs which appear

within one or more embedded clauses; examples which give rise to sloppy as well as strict

identity readings; and licit target VPs within syntactic islands, as well as within just one conjunct of

a coordinated structure. Further, an alternative analysis of the examples at issue as cases of one or

more independently dropped objects can be demonstrated to be untenable.

2. Key Aims of the Thesis, and a Chapter-by-Chapter Summary


16/133


2.1. Overview of the key aims and tasks undertaken

My principal aim in this thesis is to begin to bring the traits of non-English, V-Stranding VP

Ellipsis data to bear on the theoretical question of how ellipsis generally, and VPE in particular, is

best treated within a broad conception of post-Principles and Parameters Theory and Minimalism (see

e.g. Chomsky 1995, 2000, 2001). As discussed in Section 3 below, this is a move which has not yet

been made to my knowledge. Given space considerations, I have chosen to focus upon those issues

which I find most pressing given our current understanding of V-Stranding VP Ellipsis as reflected

in the extant literature, as well as upon issues which might prove most useful in helping to foster

further work which shares this aim. There are two principal ways in which I approach this goal.

The first task which I undertake is to make the case that sufficiently conclusive evidence now

exists to warrant data from Hebrew, Irish, and Swahili being added to the well-studied data of English

VP Ellipsis as core facts which any theory of the phonological, interpretive, and syntactic traits of VP

Ellipsis should capture. In so doing, I propose a method of diagnosis which might be employed in

future research in order to establish that (V-Stranding) VP Ellipsis is or is not instantiated in a given

language. The second core task undertaken is to begin a discussion of how the various versions of

generative syntactic VP Ellipsis theoryall designed presently to account just for English datafare

with respect to V-Stranding VPE data from Hebrew, Irish, and Swahili.

In the following two sections, I lay out the way in which these tasks are carried out, including

a chapter-by-chapter summary.

2.2. Task #1: Establishing a V-Stranding VP Ellipsis diagnosis

The first part of this thesis presents and applies steps which can reasonably be taken to be


17/133


involved in establishing a V-Stranding VP Ellipsis diagnosis in a given language. This set of

diagnostics is presented in Chapter 2, and is compiled from what one can glean from the extensive

literature on English VP Ellipsis as being among the construction's core traits, as well as from the

traits of non-VPE ellipsis types such as null objects, gapping, and stripping (the last known

alternatively as 'Bare Argument Ellipsis'), which are to be ruled out in making a VP Ellipsis diagnosis.

With respect to the content of this chapter, an initial point is of note. Since, as mentioned

above, there are currently sixteen languages for which suggested diagnoses of V-Stranding VP

Ellipsis have already been made, it might be asked whether a need still exists for the positing of

diagnostics for this construction. My answer to such a question, however, is that such a need does

indeed exist. VP Ellipsis was not generally considered to exist outside English until the early 1990s.

This means that the question of what should count as diagnostics for the construction was not a topic

subjected to the debate of experts in the field during this time, since there was no diagnosing taking

place.

This has given rise to a situation in which, once a literature on non-English VPE didbegin to

emerge, there has not been systematic use of a consistent set of traits which have been shown to hold

for the data in each language which were claimed to instantiate VPE. There are some notable

exceptions to this (see especially the careful work of Ngonyani 1996a, as well as initial work by

Doron 1990 and McCloskey 1991), but there are also several cases in which just the traits of V-

Stranding (i.e. satisfying the requirement that the null VP be governed by an overt inflectional head,

as proposed by e.g. Lobeck 1992, 1995 and Zagona 1988b) and the presence of some other

traitsloppy identity readings, for instancehave been used alone to motivate a claim of VP Ellipsis.

Such inconsistency in the application of diagnostic evidence seems likely to have contributed


18/133


to the present situation in the literature on the derivation and isomorphism requirements of VPE.

Thus, and as discussed further in Section 3 below, English has remained the sole language from which

data are typically considered in such literature, with non-English facts relegated at best to the status

of footnotes.

For this reason, it seems to me that it would be helpful at this point to propose an explicit

method by which any claim that VP Ellipsis exists in a given language can be motivated. It is my

hope that, by doing so here, such diagnostics can begin to be debated, and, if necessary, revisions and

augmentations can be proposed. By arriving at a more or less agreed-upon set of traits which VP

Ellipsis should be expected to have in any given language, future research should be better able to

make a well-motivated case that VPE exists in other languages outside those treated here, as well as

being able to rule out the existence of the construction in other languages. Ideally, this process

should thus help us as a field to deepen our understanding of the cross-linguistic core traits and points

of variation for the construction. This would enrich the empirical base from which VPE theory (and

ellipsis theory more generally) is derived, thereby advancing the development of formal treatments

of the construction toward greater cross-linguistic significance and tenability.

With this in mind, Chapter 2 is concerned with two main tasks. First, it lays out means by

which V-Stranding VPE can be distinguished from other types of null anaphora, such as Stripping,

Gapping, and Null Complement Anaphora. This is presented alongside a discussion of the relevant

traits of English VPE that V-Stranding VPE can also be reasonably expected to display. Second, this

chapter presents several case studies of careful analytic discrimination between V-Stranding VPE and

an alternative analysis involving independently elided null arguments. I view this latter issue as a

persistently under-recognized and inadequately addressed problem within the existing literature on


19/133


non-English VPE.

Using the syntactic traits of each individual language, it is shown that Irish does not allow its

direct or indirect objects to elide while other VP-internal material remains overti.e. that it has no

null object constructionbut that it does allow its such objects to drop in the sole situation in which

all other VP-internal elements elide as well. Further, it is shown that putative Irish VPE examples

involve elision of the VP-internal subjectas well, aspart ofthe elision of the entire VP. Irish subjects

are well-known to licitly drop when co-occurring with synthetic V forms which bear overt subject-V

agreement marking, and to be unable to drop when accompanied by an analytic V which bears no

such inflection. Even in the latter case, however, the Irish subject still goes missing in all putative

VPE datai.e. in data in which all other VP-internal material also surfaces as null.

In turn, Hebrew and Swahili (with accompanying discussion of the related Tanzanian Bantu

language of Ndendeule) are shown to allow only their direct objects to drop independently, and this

only under certain conditions. Hebrew null objects are shown to be obligatorily inanimate, while

animate direct objects can elide when the other VP-internal elements do. Swahili null direct objects

are shown to be licit only when co-occurring with an object-agreement morpheme on the V. Thus,

(A) using animate direct objects in Hebrew, (B) using Vs which lack an object-agreement morpheme

in Swahili, or (C) using VPs which contain more internal arguments than just a direct object in all

three of Hebrew, Swahili, and Ndendeule provide ways to rule out the possibility of an alternative

Null Object analysis for putative VPE structures in each language. When this is done, we see that

grammatical data exist in both Hebrew and Swahili which cannot be analyzed as involving null

objects, but which nonetheless involve an overt main V and all remaining VP material phonetically

null.


20/133


The final section of Chapter 2 then considers Japanese and Korean, for which I compile

arguments from various sources and perspectives to the effect that V-Stranding VP Ellipsis should

be concluded notto exist, contra the proposals in Otani and Whitman (1991). First, and unlike in the

cases of the thesis' core languages just discussed, Japanese and Korean allow each argument internal

to the VP to elide independently, and without any restrictions such as those found for null objects in

e.g. Hebrew, Swahili, and Ndendeule. Examples of putative V-Stranding VPE are thus not

distinguishable from structures in which the VP has remained intact, but with every element internal

to the VP having elided individually.

This problem then combines with an array of empirical arguments against a VPE analysis given

by Hoji (1995, 1998), Oku (1998), and Tomioka (1998) for Japanese, and by Park (1997) and Kim

(1999) for Korean. This body of work shows that the relevant Japanese and Korean data display a

range of traits which are either inconsistent with the behavior of English VPE, or which cannot be

accounted for syntactically under a VPE treatment. Taken together, these results combine to present

a strikingly solid case that Japanese and Korean should be treated as having only a null argument

possibility, and nota V-Stranding VPE construction. This conclusion thus provides a point of

contrast with the results attained for the core languages of Hebrew, Irish, and Swahili discussed

earlier in the chapter.

Chapter 2 thus aims both to demonstrate how one can go about establishing a VP Ellipsis

diagnosis in a given language, and, simultaneously, to make the case that such a VPE diagnosis is

well-motivated for the languages of Hebrew, Irish, and Swahili. In its discussion of Japanese and

Korean, it also provides a case study of data from languages for which applying the diagnostics laid

out here results in an argument that VPE should notbe diagnosed. Instead, a construction distinct


21/133


from VP Ellipsisin this case, positing the existence of independently dropped null

argumentsprovides the best analysis.

2.3. Task #2: Bringing V-Stranding VPE data to bear on the theory of VP Ellipsis

The second core part of this thesis takes as its point of departure the claim argued for in

Chapter 2, namely that Hebrew, Irish, and Swahili have a construction which is best treated as a case

of V-Stranding VP Ellipsis. This means, in turn, that V-Stranding VP Ellipsis should be viewed as

a bone fide VPE sub-type, involving data which form a natural class with those of the familiar English

Aux-Stranding variety. The desideratum, then, is that the V-Stranding data from the core languages

treated here be derived in the same basic way as is English VP Ellipsis. The task of arriving at a

unified analysis of V-Stranding and Aux-Stranding VPE is thus the main focus of Chapters 3 and 4.

Specifically, these chapters explore the relative treatments given to V-Stranding VPE by the

two key competing syntactic analyses of English VP Ellipsis in the literatureanalyses which, in

some form, have formed the focus of an analytical debate which has endured throughout most of the

history of generative ellipsis theory. I argue that the PF Deletion view of ellipsis (for English VPE,

see e.g. Tancredi 1992, Fox 2000, and Merchant 2001, reaching back to Hankamer and Sag 1976,

Sag 1976 and other work), allows a markedly more natural and parsimonious account of V-Stranding

VPE data than does theLF Copying approach (for English VPE, see e.g. Hardt 1992 et seq. and

Lobeck 1992, 1995 et seq., reaching back to earlier treatments by Williams 1977 and others). Given

the assumption that V-Stranding VPE and English VPE should receive a unified treatment, this result

provides a strong argument in favor of the adoption of a PF Deletion approachand against adopting

LF Copyingfor VP Ellipsis as a whole, and perhaps for ellipsis in general.


22/133


Arriving at this result involves addressing two main questions, each of which forms the

respective focus of Chapters 3 and 4: the first involves the position and phonetically overt nature of

the target-clause main V, and the second involves the identity requirements between this V and its

antecedent-clause counterpart. As laid out in Chapter 3's discussion of the V-Stranding effect, the PF

Deletion view requires the target clause VP of V-Stranding VPE to be base-generated with fully

fleshed-out internal syntactic structure; the VP is then deleted or not pronounced at PF, so long as

certain semantic parallelism and identity constraints between the antecedent and target VPs are

satisfied. Here, then, the target clause main V can be base-generated internal to the VP, just as in

ordinary non-ellipsis examples. The V will subsequently raise out of the VP at some point prior to

that at which the ellipsis effect takes place, allowing a derivation much like that of English VPE, with

the addition just of the fact of main V raising to Infl.

In contrast, the LF Copying approach base-generates the target clause's superficially null VP

without internal structureoften, as in the work of Hardt and Lobeck, as a null proform

elementwhich then receives its interpretation via the copying in of its antecedent VP's

representation at LF. This means that the superficially overttarget clause main V must be base-

generated outside the null VP. Under LF Copying, then, the basic structure of V-Stranding VPE

target clauses in a given language will be quite different from that of non-VPE examples in the same

language, with the main V generated internal to the VP (under normal assumptions) for the non-

ellipsis cases, but external to the VP in ellipsis cases.

In Chapter 4, each of PF Deletion and LF Copying are then considered with respect to the

question of what the semantic isomorphism requirements are which hold between the antecedent and

target clauses of V-Stranding VPE, with respect to the main V of each. The question which arises


23/133


at this point is whether this main V is interpretedin this VP-external position as well. Although the

semantic effects of verbal head movement are not easily probed, it turns out that V-Stranding VPE

provides a way to do just this. On the one hand, it is possible that the raised main Vs of the

antecedent and target clauses are part of what must be isomorphic between the antecedent and target

VPs; this situation is tacitly true for English VPE, given that the target-clause main V there is

necessarily null, and so is necessarily understood as identical to the antecedent-clause main V:

(6) Leila might like to wear hats, and Madeline might [like to wear hats] too.

Examples like (6) are only licit when the null VP of the target clause is interpreted as having the same

core meaning as that of its antecedent clause. Were the Vs of these VPs to be non-identical (with the

exception just of certain allowable differences in the inflection of the two Vs, as has been known

since at least Sag (1976)), the example would be illicit, because there would be no way to recover

the different meaning of the target-clause main V as opposed to the antecedent V:

(7) *Leila might like to wear hats, and/but Madeline might [hate to wear hats] (too).

For V-Stranding VPE, in contrast, it is an a priori possibility that the main Vs of the two

clauses might notneed to be the same, since both Vs are overt. Were this to be the case, the VP-

external surface position of the antecedent- and target-clause Vs would essentially have not only

phonetic import, but semantic import as well. This is an empirical question, the result of which

requires resolution in order for an analysis of V-Stranding VPE structures to be developed: it is

essential to know whether the account adopted needs to include or exclude the main Vs of the two

clauses in the material which it holds parallel. Claims that V-Stranding VPE does or does not require

such verbal identity have been made at various points in recent literature (see e.g., respectively,

Doron 1990, and Potsdam 1997 versus Otani and Whitman 1991, Lasnik 1997, and Doron 1999),


24/133


4Efforts were made when working with native speakers to ensure that the pragmatics of these examples were notresponsible for the ungrammaticality observed. In addition, note that, in the case of non-identical binyanim inexamples like (9), the argument structure of the antecedent and target VPs necessarily changes as well. In this

example, for instance, the antecedent VP has a single internal argument, the dative le-Tel Aviv 'to Tel Aviv', whilethe target VP includes the meaning of an additional accusative 'ota 'her'. As noted in Chapter 4, this is clearly partof what is responsible for the ungrammaticality of such examples, along with the binyan mismatch.

On a larger level, all data in this thesis which are not attributed to an explicit source were gathered in my ownwork with native speakers of the language, all of whom spoke Hebrew as their primary language through at leastgraduation from high school, and most of whom were aged in their early 20s during the time of my work with them.Though all resided in Canada at the time of this work, a majority (A) had been living outside Israel for no more thana few years, and (B) continued to speak Hebrew in some aspects of their daily lives. My work with the data involvedhas also been facilitated by the following particularly helpful reference works on Modern Hebrew: Donnet-Guez(1995), Bolozky (1996), Dahan (1997), and Doniach and Kahane (1996).

but each such extant claim involves data which are either ambiguous between null object and V-

Stranding VPE analyses, or otherwise involve languages in which a clear V-Stranding VPE analysis

has yet to be established.

In answering this question, Chapter 4 presents a novel empirical paradigm from Hebrew. As

a Semitic language which is now (given the results of Chapter 2) been established as a clear V-

Stranding VPE language, Hebrew is an especially good choice for investigating the question. This

is because its rich derivational morphologyinstantiated as its conjugation paradigms known as

binyanimallows various senses of verbal identity to be probed. The result in Hebrew turns out to

be that the main Vs of the antecedent- and target-clauses must be identical in both their root andin

their derivational morphology. V-Stranding VPE is thus ungrammatical when both the root and

binyan of the two Vs differ, when the roots differ but the binyan is held constant, and when the

binyanim differ but the roots are held constant. Just as is true of English VPE, however, the

inflection of the two Vswhich, in Hebrew, will be either or both of tense and subject-V

agreementmay vary. (8-10) provide illustrative examples:4


25/133


Hebrew: *VPE with Same V Binyan (hif'ilCausative), Different V Roots (Antec Nun-

Samex-Ayin, Target Bet-Aleph)(8) Q: Rivka hisi'a otax le-beit ha-sefer?

Rivka drive[Past3Fsg] ACC.you[Fsg] to-house the-book

'(Did) Rivka drive you to school?'

A: (Ken,) hi *hevi'a / hisi'a.yes she bring[Past3Fsg] drive[Past3Fsg]

'(Yes,) she *brought/drove [me to school].'(cf. English 'Yes, she brought me.')

Hebrew: *VPE with Same V Roots (Nun-Samex-Ayin), Different V Binyanim (Antecpa'al

Simple, Targethif'ilCausative)

(9) Q: Li'ora nas'a etmol le-Tel Aviv?Liora travel[Past3Fsg] yesterday to-Tel Aviv'(Did) Liora travel yesterday to Tel Aviv?'

A: *Ken hisa'ti *(ota).yes drove[Past1sg] ACC.her

'YesI drove (lit. 'caused to travel') .'(cf. English 'YesI drove her.')

Hebrew: VPE with Same V Root and Binyan, Different Tense and Subject-V Inflection(10) Q: Tazmini et Dvora la-mesiba?

invite[Fut2Fsg] ACC Dvora to.the-party'(Will) (you) invite Dvora to the party?'

A: Kvar hizmanti.already invite[Past1sg]

'(I) already invited [Dvora to the party].'

Facts parallel to those presented here for Hebrew hold in Irish V-Stranding VPE as well, and

the requirement also holdsif uninterestinglyfor the overt versus null antecedent-target V pairs

of English VPE. In all three, then, the inflection of the antecedent- and target-clause main Vs may

generally vary licitly, but the actual V roots and their derivational morphology may not. I term this

generalization the Verbal Identity Requirement, and hypothesize that it is a cross-linguistic trait of

the VP Ellipsis construction.


26/133


Although the effects of the Verbal Identity Requirement in English are captured under either

a PF Deletion or an LF Copying account, its effects in V-Stranding VPE turn out to have significant

consequences for the relative success of the two lines of analysis. Under PF Deletion, parallelism and

identity between the antecedent and target VPs is achieved via a set of constraints which must hold

of the two VPs' LFs in order for deletion or nonpronunciation of the target VP to be allowed (or for

such omission of the phonetic content of these VPs not to cause the derivation to crash).

Generalizing over the various instantiations of this line of analysis, these isomorphism constraints can

be taken to require essentiallymodulo certain focus effectsmutual entailment between the

antecedent and target VPs, once these VPs are type-shifted to a proposition via a process such as

Existential Type Shifting, as in Schwarzschild (1999).

I argue that the facts of Verbal Identity in V-Stranding VPE will be captured under this

conception of PF Deletion accounts, so long as the main Vs of V Raising languages are interpreted

internal to the VP. When this is true, the semantic content of these Vsincluding their derivational

morphology, but crucially not their inflectionwill be correctly held identical as part of the semantic

mutual entailment requirement already needed to capture the facts of English VPE. This can be done

either by relegating the V-to-Infl movement of such Vs to PF (as in many recent proposals; see e.g.

Chomsky 1995, 2000, 2001; Boeckx and Stjepanovic 2001), or, as I argue here, via obligatory

reconstruction of Vs to their base, VP-internal position, at LF. This, then, is the only additional

assumption which must be added to the mechanisms already in place under PF Deletion to capture

the facts of V-Stranding VPE along with those of English VPE.

Part of the fundamental difference in approach between PF Deletion and LF Copying lies in

how the two enforce semantic isomorphism between ellipsis antecedent and target clauses. For LF


27/133


Copying, such identity comes as a natural result of the copying mechanism itself: the target VP has

no inherent content, and receives this content only via the copying in of the semantic representation

of the antecedent VP. What is needed in order for such copying to capture the Verbal Identity

Requirement of V-Stranding VPE is thus for the antecedent main V to be part of the semantic

material copied into the target clause.

It might seem initially that this could be achieved using the same treatments just described for

PF Deletionnamely, by either relegating the movement of the antecedent main V to PF, so that it

lies internal to the VP prior to and into LF, or by raising this V prior to LF, but then reconstructing

it back into the VP at LF. However, this does not work as straightforwardly for LF Copying. As

described above, it is necessary under LF Copying to base-generate this target V external to the VP,

in a (sub-)head of Infl, in order to capture the fact of its phonetic overtness in V-Stranding VPE

target clauses. If the antecedent clause VP's representation is then copied into the null VP of the

target clause at LF, the target clause's distinct VP-external main V, lying in Infl, would presumably

then need to be linked up with the newly copied-in VP-internal main V from the antecedent clause.

At this point, added difficulties for LF Copying ensue. Although identity in interpretation

between the antecedent and target clause VPs should already have been taken care of via just the LF-

copying act itself, we see that the Verbal Identity issue still needs to be dealt with. Specifically,

capturing the facts of the Verbal Identity Requirement requires that the overt, base-generated main

V inInfl of the target clause be held semantically isomorphic to the newly copied-in VP-internal main

V from the antecedent clause. This will require a new mechanismone which, on the most workable

assumptions, will essentially replicate the isomorphism constraints of the PF Deletion approach in a

way specific to the antecedent- and target-clause main Vs specifically.


28/133


The need for this additional isomorphism constraint, however, is more than just a mere

stipulation required to make the account work. The most fundamental claim of the LF Copying

approach is that the target VP of ellipsis has no internal structure or inherent content, and that it

receives this content entirely from its antecedent VP. In order to allow LF Copying to extend its

range to encompass V-Stranding as well as English VPE, however, we see that LF Copying must be

modified to also incorporate the semantic isomorphism constraints which hold between material of

the antecedent and target VPs within the PF Deletion accountover and above the copying

mechanism already in place, which on its own was intended to achieve the entirety of such

isomorphism requirements.

For these reasons, then, the contents of Chapters 3 and 4 argue that LF Copying be rejected

as a viable way in which to capture the facts of both the Aux-Stranding VPE of English, and of the

V-Stranding VPE varieties outside Englisharguing instead that PF Deletion provides a superior

account. PF Deletion is seen to capture the full range of facts from V-Stranding VPE

straightforwardly, using just the mechanisms already in place to treat English VPE, and with just the

additional reasonable assumption that raised Vs which are pronounced external to the VP are

nonetheless interpreted in their base position, internal to the VP.

LF Copying, in marked contrast, requires modification of the phrase structure of V-Stranding

VPE target clauses in anomalous ways, and has the same need as PF Deletion to interpret raised main

Vs internal to the VP. Even with such assumptions, LF Copying cannot capture the Verbal Identity

Requirement of V-Stranding VPE without adopting the same sorts of semantic comparison metrics

between the antecedent and target clauses which are already needed under the PF Deletion

approachmetrics whose presence, when taken seriously, undermines the basic spirit and


29/133


fundamental tenets of LF Copying.

Chapter 5 of the thesis, finally, presents a summary and concluding remarks. This includes a

discussion of those issues which, it seems to me, present the most interesting next steps which future

research might take. It also mentions the references of which I am presently aware which posit that

VP Ellipsis exists in other languages which could not be discussed here: Brazilian Portuguese, Serbo-

Croatian, Russian, Finnish, Hungarian, Basque, Mandarin Chinese, and Tagalog. Among the sixteen

languages outside English for which I am aware of VPE claims having been made, I take the VPE

claims for data in the eight languages just listed to be neither conclusively motivated nor conclusively

refuted at this point, and thus in need of further work toward a clear diagnosis.

3. On the Motivation for Integrating V-Stranding VP Ellipsis Data into (VP) Ellipsis Theory

Before moving on to Chapter 2, I would like to conclude this introduction by providing some

initial discussion of what I see as the underlying motivation for undertaking the present study. The

compelling nature of this motivation will hopefully become increasingly clear to the reader as the

thesis proceeds.

As far as I am aware, the work which has been done to date on VP Ellipsis can be considered

to fall into one of two distinct groups. On the one hand, a prolific literature on the analytic treatment

or derivation of what has become an established set of English VP Ellipsis data has spanned nearly

the entire history of generative grammar itself. This includes early work in the 1970s, by Sag,

Williams, Postal, Hankamer, and many others, on the construction's basic traits, and on the debate

between a deletion versus pronominal analysis. Discussions in the 1980s and early 1990s, principally

by Lobeck, Zagona, and Chao, expanded the early, seminal work on the construction, focusing at this


30/133


5As discussed in the concluding Chapter 5, a complementary set of non-English VP Ellipsis claims also exists (forlanguages like Mandarin Chinese, Moroccan Arabic, and Serbo-Croatian) involving anAux-Stranding variety of theconstruction. Although these data are not the focus of the present study, the diagnostics laid out in Chapter 2 will, onthe whole, be directly applicable to the question of whether or not they ultimately constitute cases of VPE.

point on the null VP's need to be properly governed by an inflectional head.

Work then begun in the early 1990s, and still continuing at present, has involved research by

Tancredi, Rooth, Fox, and Merchant, among others, on correlations with phonological deaccenting,

and on how these relate to the parallelism and identity constraints which hold between the antecedent-

and target-clauses. Work on LF Copying approaches has also continued, as in work by Hardt (1992

et seq.), Frazier and Clifton (2001), and others.

Beginning roughly in the early 1990s, a parallel body of work has developed which has claimed

that VP Ellipsis exists in languages outside English as well, but that such non-English instantiations

involve the main V obligatorily remaining overt.5 A good portion of this work, however, has yet to

be accessed or discussed by much of the larger linguistic community so far; this sense is culled from

my own personal communication with other linguists, as well as from my reading of the literature on

VP Ellipsis theory described just above.

This body of work was begun in an early manuscript by Doron (1990) on Hebrew, in work by

McCloskey (1991) on Irish, by Laka (1990) on Basque, and by Otani and Whitman (1991) on

Japanese, Korean, and Chinese (with suggestions made as well for 'Portuguese', without reference

to a particular dialect). The basic notion that VPE of a V-Stranding variety might exist outside

English may have its origins in earlier discussions by Huang (1987b, 1988, 1991) and by Chao (1987).


31/133


6In this body of work, the argument that V-Stranding VPE exists in the particular language being studied is oftenused as part of a larger theoretical proposal that the main V lies in a suitably high position within the language's phrasestructural architecture to escape the elision of its verb phrase.

Such work, thenthough also pursuing an analytically-oriented goal6can be contrasted with the

body of work on English VPE just described in being concerned principally with establishing whether

or not certain non-English data might be able to be analyzed as forming a natural class with the

established set of English VPE data.

Since this initial research, claims that V-Stranding VP Ellipsis exists have subsequently been

made for Spanish and Italian (Lpez 1994 and subsequent work), European Portuguese (Martins

1994, 2000), the Southern Bantu languages Swahili and Ndendeule (Ngonyani 1996a,b, 1998),

Serbo-Croatian (Stjepanovic 1997a,b and subsequent work), Russian (with cursory discussion of

Polish and Czech as well; McShane 2000), Finnish (Holmberg 1999, 2001), Hungarian (Bnrti 1994,

2001, Bartos 2000, 2001; Gyuris 2001), and Tagalog (Richards 2002).

Further work on the initial languages for which the construction was first claimed to exist has

continued during this time as well, including some work which has argued against a VPE analysis for

the data at issue in certain languages (as in Hoji 1995, 1998, Tomioka 1998, and Oku 1998 for

Japanese; Park 1997 and Kim 1999 for Korean; and Depiante 2000, 2001 for Spanish). For other

languages, subsequent research has worked to further substantiate the initial claim that VPE is

instantiated (as for Hebrew, e.g. Ussushkin/Sherman 1997, Doron 1999, Goldberg 1998, 2002a,b;

and for Irish, e.g. McCloskey 1995).

These two lines of analysisone developing and debating the derivational treatment of English

VPE, and a second investigating the tenability of VPE diagnoses for languages outside Englishhave

developed in tandem, but with no significant intersection, throughout the 1990s and to the present.


32/133


This is understandable from both sides of the picture. From the standpoint of research on VP Ellipsis

outside English, it has long been assumed that VP Ellipsis is a construction relegated solely to

Englishdue, presumably, to the cross-linguistic rarity of its sort of auxiliary V system, in which

tense and aspect are commonly marked on words distinct from the main V. Thus, since the possibility

ofV-stranding VP Ellipsis in V Raising languages had not been established or considered previously,

the 1990s literature cited above on this construction has, quite rightly, focused just on arguing that

the data at issue do indeed instantiate VP Ellipsis, having the same core empirical traits as the English

construction.

From the opposite perspective, it is equally reasonable that the literature on the derivation and

isomorphism requirements of VP Ellipsis has excluded such non-English data from consideration

during this time. As alluded to above, some of the claims positing non-English VPE in certain

languages have seemed lacking in full motivation, thus being in need of further study to arrive at a

more conclusive diagnosis. It would be wrong to essentially contaminate the theory of VP Ellipsis

by expanding it too hastily to encompass data which might not, in fact, form a natural class with the

established data from English.

For these and other reasons, then, such work often mentions non-English V-Stranding VPE

only in a footnote or an aside, usually citing only the work of McCloskey and/or Doron. Moreover,

almost without exception, such data are not incorporated into the core examples treated in the main

text (see e.g. Fiengo and May 1994:p.18, fn.148, and p.161, fn.23; Lobeck 1999:p.19, fn.14;

Merchant 2001:p.71; and Johnson 2001:p.175, fn.11). Further, much of the work on VP Ellipsis

outside English has not been widely distributed, and so is likely not to have been available to

researchers focusing on the types of analytical issues which have been explored for English VP


33/133


Ellipsis.

Nonetheless, it has always been a mystery why VP Ellipsis should be such a robust trait of

English, but should be entirely absent in other languages. This is all the more true given that other

types of null anaphoraincluding Sluicing, Gapping, and Stripping (also known as Bare Argument

Ellipsis)seem to be attested in nearly every language which has been studied. The main

contribution of this thesis in this light, thenand particularly in terms of the content of Chapter

2might thus be seen as helping to bring out the fact that we do appear to be making progress on

solving this mystery. It is my hope that the present content may aid future research in helping to add

other V-Raising languages to the natural class of VP Ellipsis languages of English, Hebrew, Irish, and

Swahili which emerges here.


34/133

23

Chapter 2

Establishing a Diagnosis of V-Stranding VP Ellipsis

0. Introduction and Chapter Overview

This chapter focuses on the issue of how to achieve a diagnosis of V-Stranding VP Ellipsis.

The approach taken here is that it is reasonable to diagnose VPE for putative V-Stranding VPE data

when two principal tasks have been accomplished. First, it should be established that the data cannot

be analyzed alternatively as other imaginable types of null anaphora, such as Stripping (Bare

Argument Ellipsis), Gapping, Null Complement Anaphora, or, especially, as a string of one or more

individually elided null arguments.

Second, it should be established that the data display those key behavioral traits of English

VPE that this thesis argues are diagnostics of VPE in general. Such traits include the ability for the

target clause to appear within one or more levels of sentential embedding, to appear within a

syntactic island (where the antecedent clause lies outside this island), and to appear within just one

conjunct of a coordinate structure (where, again, the antecedent clause lies outside the entire

coordinated structure).

For the core languages under consideration in this thesisIrish, Swahili, and Hebrewthis

diagnostic work of both steps has already been carried out in relatively complete fashion, and such

work will be outlined briefly in Section 1 of this chapter. However, there is one significant

exception to this in the extant literature on particularly Hebrew and Irishand this concerns the

question of analyzing such data as the product of one or more independently elided null arguments.

Careful demonstration that this remaining alternative analysis is not possible for the data claimed


35/133

Goldberg, Chapter 2 Establishing a Diagnosis of V-Stranding VPE 24

to involve VPE in each of these languages is thus the concern of the bulk of this chapter.

Such ambiguity between a null argument and V-Stranding VPE analysis has been a

persistently under-recognized problem which has plagued the literature on V-Stranding VP Ellipsis

(and literature on null objects as well). Most typically, the specific ambiguity in the putative V-

Stranding VPE data is with a null direct object, since transitive Vs are often used in the examples

involved. Such surface-string ambiguity is not an issue forEnglish VP Ellipsis, since its main V

necessarily elidessomething which would not happen in a null argument structure.

For languages such as those considered here, however, the main V lies overt within Inflthe

inflectional domain of the sentenceso that surface-strings in which all VP-internal constituents are

null are inherently ambiguous between a V-Stranding VP Ellipsis analysis, and an alternative

analysis in which the VP remains intact, but each of its internal constituents has dropped

independently. A principal focus of the present chapter will thus be on how to develop and use

control methods for null arguments in putative V-Stranding VPE data. As discussion of this

proceeds through the issues specific to each of Hebrew, Swahili, and Irish, we will see essentially

three case studies of languages for which a diagnosis of V-Stranding VPE (as opposed to one

positing null arguments or sorts of null anaphora) turns out to offer the best characterization

available for a given set of data.

This chapter is organized as follows. Section 1 gives an overview of diagnostic tests other

than those involving null arguments to be used in determining the presence of VP Ellipsis. Section

2 then turns to the null argument-VPE ambiguity problem, laying out the details of how such

ambiguity arises, and accompanying these with initial steps which can be taken to rule out a null

argument analysis for putative VP Ellipsis examples in any given language. Sections 3 through 5


36/133


then show how this can be done for the core languages focused upon in this thesis, crucially

employing independent properties of the particular language in each case.

Specifically, Section 3 presents the results of my own previous work on Hebrew (Goldberg

2002a,b), in which it is shown that, while Hebrewlike Swahili and Ndendeuledoes have a null

direct object construction, it is restricted just to those direct objects which are inanimate. Further,

it is shown that VP-internal elements besides the direct object are not able to elide independently (i.e.

without all VP-internal elements eliding) in this language. Putative Hebrew VPE examples are thus

shown not to be able to be analyzed alternatively as cases of Null Objects, since such VPE examples

can elide not only the direct object, but additional internal arguments as well, including those direct

objects which are animate. Thus, the use of animate direct objects, as well as of VPs in which the

direct object is not the sole internal argument, can be employed to rule out a Null Object analysis for

putative VP Ellipsis examples in the language.

Section 4 discusses this issue for Swahili and Ndendeule, using the results of the careful work

of Ngonyani (1996a, 1998) for these languages, which unfortunately has not yet received wide

distribution. We see in section 4 that, like Hebrew, Swahili and Ndendeule do allow direct objects

to surface independently as null, with other VP-internal elements overt, but do not allow this for

other VP-internal elements. For these languages, Ngonyani's work shows that such Null Objects in

these languages are licit only when accompanied by an object-agreement morpheme which appears

on the V. Grammatical examples in these languages in which all VP-internal material surfaces as

null, however, lack this object-agreement morpheme. This, along with the ability of a wide range

of verbal argument sequences to elideincluding both the applied and direct object of applicative

constructions, the causee and lower object of causative constructions, and infinitival complements,


37/133


none of which correspond to an extant object-agreement morphemedefinitively rules out a null

object treatment for the putative VP Ellipsis examples in these languages. Thus, constructing

examples in these languages in which an object-V agreement morpheme is not present will provide

a way to rule out a Null Object analysis when testing putative instances of VPE.

Section 5 then discusses the case of Irish. Although McCloskey (1991, 1995) did not

explicitly rule out a null argument analysis for the data which he argued to instantiate V-Stranding

VPE, I show here that such data cannot be analyzed alternatively as involving independently elided

arguments within an intact VP. The basic sentential order in finite clauses is VSO in Irish, and the

subject of such clauses has been argued extensively by McCloskey to remain internal to the verb-

phrasal domain in surface representations. It is shown here that, from what is already known about

Irish syntax, some arguments may elide in this language, but only in the presence of agreement

morphology on a verbal, P, or N head.

Since Irish Vs do not agree with direct objects, such objects are not able to surface

independently as null. Further, the verbal paradigms of the language are typically mixed between

V forms standardly referred to as 'synthetic', which bear subject-agreement morphology and so allow

(and in fact require) null subjects, and V forms standardly referred to as 'analytic', which do not show

agreement with the subject, and so require that their subject be overt. Therefore, putative VPE

examples in this language in which the direct object elides, as well as those in which a VP-internal

null subject is accompanied by analytic V forms which lack subject-agreement, cannot be cases of

independently elided VP-internal direct objects or subjects. In turn, using such examples when

testing V-Stranding VPE in this language will rule out the possibility of an alternative null argument

analysis.


38/133


Finally, Section 6 presents a case study of a set of data from the literature on Japanese and

Korean which, like the data from the core languages discussed earlier in the chapter, are again

ambiguous between a null argument and V-Stranding VPE analysis. Unlike the earlier core cases,

however, the Japanese and Korean data are best analyzed as involving one or more individually

elided null arguments, and notas V-Stranding VPE. This section thus provides a compilation of

arguments to this effectmany of which come from work which has gone unnoticed in much recent

literature.

Section 7 then closes the chapter with some concluding remarks.

1. A First Consideration of VPE Diagnostics

Because the presence of VP Ellipsis has not previously been explored outside the first

language in which it was observed (namely English), we have lacked a set of established diagnostics

for the phenomenoncontrasting with the situation for e.g. unaccusativity, negative polarity, and

so on. It can nonetheless be noted that English VP Ellipsis has a characteristic set of behavioral

traits, the confluence of which is not found in other types of null anaphora. In beginning to consider

putative V-Stranding VPE data, then, it is reasonable to posit that a majority of the traits

characteristic of English VPE should hold of the putative V-Stranding VPE data as well, and further

that none of the traits should be absent without a clear reason adduced from independent factors.

1.1. Licensing by an inflectional head

English VP Ellipsis is most easily distinguished from other types of ellipsis in the language

by its requirement that one or more Auxiliary Vs be overt, while the entire contents of the verb


39/133


phrase, including its main V, appear as null:

(1) Bruce fed Colin some yogurt, and Bonnie did [VP

feed Colin some yogurt] too.

Here, the pleonastic Auxiliary do is stranded; modals may also be stranded, as well as instances of

perfective have, progressive be, and passive be, with or without the presence of other Auxiliaries or

negation, in most cases:

(2) Mikael had arrived by noon, but Hiro hadn't [VP arrived by noon].

(3) Joey is/might be bringing spanikopita to the party, but Becca isn't/might not

be [VP bringing spanikopita to the party].

(4) The mushrooms were (being) fried up by Laurie, and the artichokes were[VP (being) fried up by Laurie] too.

Some restrictions do hold for English VPE in terms of what may licitly serve as the stranded element

preceeding the null VP, concerning especially instances ofbe, of nonfinite to, and of negation; see

e.g. Johnson (2001) for a recent review of some of the relevant issues and references.

I will abstract away from some of the finer details of this generalization for English VPE here,

keeping instead to the present focus of using English VP Ellipsis as a source of comparison for

claims of VPE in other languages. What is crucial for this purpose is that the various accounts for

such facts converge in the sense of accounting for this fact by implementing formally the

generalization that the null VP of VP Ellipsis (as well as the null IP and NP of Sluicing and NP

Ellipsis, as described below) must occur in the presence of an inflection-bearing head located in the

Infl domain. Thus, all such accounts work to rule out ungrammatical examples of English VPE like

the following, which contrast with the licit sentences of (1-4) above in lacking the presence of such

a head: in (5), inflection lies on the main V, which is not in Infl, and is, furthermore, elided; in (6),

the stranded nonfinite be which (presumably) lies in Infl does not bear tense or agreement; and, in


40/133


1These examples are constructed so as to rule out other types of ellipsis which are ungrammatical in VPEenvironments. For instance, the target clause of (5) would be licit as an instance of the distinct construction of Strippingif it were not embedded, and that of (7) be licit as a case of Null Complement Anaphora if it did not precede itsantecedent clause in this 'backward ellipsis' configuration.

2For readers who may not be familiar with this body of work, Sluicing (first described by Ross 1969) can becharacterized as the elision of an IP complement to an interrogative C head, and NP Ellipsis (under the DP hypothesisof noun phrases) as the elision of an NP complement to a possessive D head. They are thus like VPE in involving theellipsis of a constituent which serves as complement to a functional head, but differ from it in necessarily involving aspecifier which is filled, as seen in the following respective examples:

Sluicing:(i) [CP [IP-ANTEC That squirrel just ate something]], but I don't know [CP what/where/why [IP that squirrel just ate (it)]].NP Ellipsis:(ii) Henry remembered to bring [DP his own [NP-ANTEC jacket]], but I had to borrow [DP Robey's [NP jacket]].

(7), the inflected main V is left overt, but, given that this is English, does not lie in Infl: 1

(5) *The administration recommends that the teachers arrive on time, and that the students [VParrive on time] too. (Lobeck 1999:ex.48b)

(6) The company asks that employees be finished with lunch by 2pm....*The company recommends that they be [VP finished with lunch by 2pm] in order to be backon the job by 2:05. (Lobeck 1999:ex.49)

(7) *Before John started [VP reading War and Peace], Mary finished reading War and Peace.(Lobeck 1999:ex.42a)

Let us first discuss the various formulations of this licensing requirement, and then consider its

general implications for V-Stranding VP Ellipsis.

Initial accounts for this generalization by Zagona (1982, 1988a,b), Lobeck (e.g. 1986a,b,

1991, 1992, 1995), and Chao (1987) (see also Contreras 1989 for what I will refer to as NP Ellipsis,

known previously as 'N-bar Deletion') appeared within a Government and Binding framework. This

body of research has generally proposed that the elided material of VP Ellipsis, Sluicing, and NP

Ellipsis

2

(which have been motivated extensively to form a natural class among the ellipsisconstructions; see especially Lobeck 1991 and 1995, as well as Chao 1987 and Johnson 2001) must

be properly governed by an inflectional head in order to be licensed. Lobeck's (1995) formulation


41/133


3This final version of Lobeck's (1995) licensing requirement is first given, to my knowledge, on page 20 of thatwork. However, the formulation is repeated verbatim several times throughout the book, including on page 52whichis the page reference given to it in Lobeck (1999).

is widely cited, and is given as follows:3

(8) Licensing and Identification ofpro: (Lobeck 1995:20)An empty, non-arbitrary pronominal must be properly head-governed, and governed by an Xspecified for strong agreement.

The null constituent in each of the three ellipsis constructions is argued by Lobeck, Zagona,

and Chao (as well as by Hardt, e.g. 1992, 1993) to involve a non-arbitrary null pronominal element,

which I will refer to aspro for present purposes. The requirement given is therefore characterized

in terms of the licensing ofnull pronominal elements. Being specified for strong agreement in

Lobeck's terms means that the X heador a head with which it agreesshows overt morphological

agreement (thus realizing features which it shares with those of another head or phrase under

government) in a productive number of cases. The requirement, under this view, will thus ensure

that the null VP, NP, and IP of each construction will be properly governed by an agreeing Infl, D,

or C head, respectively. For VP Ellipsis, Lobeck proposes that the relevant type of agreement feature

at issue is [+tense], which will be realized on a verbal head, but is initially located on a sub-head of

Infl.

Various Minimalism-consistent reconceptions of this inflectional licensing constraint are

possible which avoid the notion of government and thus of proper government (see e.g. Merchant

2001:60-61, 2004; Potsdam 1998:77). The core insight from all such formulations, however, is what

is important here: VP Ellipsis is considered to be licensed only when it occurs in the presence of a

local, tense-inflected head. As we will see below, this trait crucially distinguishes VPE from other

anaphoric constructions involving null material, such as Gapping and Stripping.


42/133


Further, this inflectional licensing constraint makes the prediction that V Raising languages

fit in to the paradigmatic requirements shown for English VPE. This is so because the licensing

inflectional head is not restricted to being an Auxiliary V, leaving open the possibility that the raised

main Vof a V Raising language might fill the same licensing role as Lobeck captured for English

Auxiliary Vs.

1.2. Additional traits characteristic of English VPE

Apart from the question of what is stranded in the target clause, a variety of other traits can

be noted to distinguish English VPE from other types of null anaphora. First, unlike a construction

such as Null Complement Anaphorawhich, though it strands a main V, only elides constituents

which express a propositionEnglish VPE is grammatical regardless of the internal content of the

VP which elides:

VP Ellipsis(9) Q: a. Did Heather know that Cinema du Parc had opened?

b. Did Heather see Seonaid?c. Did Heather walk to the department?d. Did Heather send the bagels over to Naokos?e. Did Heather put the bagels on the counter?f. Did Heather tell you that Jupiter Room is open?g. Did Heather tell you to come over?

A: She did.

(Putative) Null Complement Anaphora(10) Q: Does Heather know that Cinema du Parc is open?

A: She knows.

(11) Q: Does Heather see Seonaid?A: *She sees.

Additional behavioral traits of English VPE include the ability of the target clause to appear


43/133


within a syntactic island (of which the antecedent clause lies outside), within just one conjunct of

a non-across-the-board coordinate structure (again with the antecedent clause lying entirely outside

the coordinated structure)both of which are impossible for null arguments involving movement

dependenciesand, especially, the ability for the target clause to appear licitly within one or more

levels of sentential embedding. Systematic grammaticality acrossallsuch environmentscombined

with a freedom to appear in a range of discourse environments without bei

goldberg phd 1st half

Documents